Oooh, did someone rattle the IBM marketting cage? Looks like IBM have realised the same as hp did a long while ago - enterprise UNIX is now a two-horse race, Power and Itanium. And it looks like the IBMers are coming out swinging wildly.
RE: tukwila and nehalem
".....when Tukwila was canceled...." Tukzila wasn't cancelled, only delayed. Now, please try and pretend every Power chip has come in right on the originally announced deadline. Just for our amusement. Do you need a little reminder that Power6 was originally slated for 2006? Then explain how Nehalem, which won't scale above eight sockets even when we get the EX chips, was able to "crush" Montvale? What application were you using? If it only needed scale of less than eight sockets and ran on both Itanium and Xeon then it would have to be Linux or Windows - the former is a complete recompile on Power and the latter doesn't run there at all. And anything that low down the enterprise scale would run faster on Xeon than Power anyway. I detect more than a faint aroma of male bovine manure!
RE: can I ask a silly question?
There are no silly questions. The simple answer is that Itanium scales much higher than Xeon can currently in a single system image. That is the UNIX SMP market where Power needs to score but where Itanium and hp-ux are winning the big deals. A new but growing market segment, large SMP instances of Windows, is also an area where Power's inability to run a variety of OSs is starting to hurt. There is pressure from Windows shops to run large instances of MS SQL, and IBM simply can't do that on Power or Xseries. For large MS SQL instances the only answer is still Itanium, and will be even when we get 8-socket Nehalem EX servers as they still won't scale as high as Tukzilla in a single Windows instance. Oh, and those 8-socket Nehalem EX boxes will be from hp - has IBM even said they will make an EX server larger than four sockets? And if they do, the real laugh will be they will need to use "node controllers" to glue two four-socket Nehalem EX motherboards into one system - derived from Itanium's cell-based backplane switch technology!
RE: Why would people upgrade to this?
"....there are still some very niche areas where a very well coded app will perform better on Itanium than other platforms....." Yes, like these three; the largest SAP installation in the World, the largest telecoms billing systems in Europe, and the largest RAC instance in Europe. All of which I understand to have been won in competition with Power and AIX. Doesn't sound so niche to me!
".....the ports may require a completely new (and expensive) set of licences....." Hilarious! IBMers trying to push the "double-the-cores-double-the-licences-only-happens-on-Itanium" bilge still? So how many core is Power going to? From four to eight, so doubling the licence costs if you buy exactly the same number of sockets. So Power is equally non-bad in this complete non-argument from the IBMers. It's telling that this is the best the IBMers can come up with! But if you want to look at this further, if a current Integrity user needs the same number of cores then Tukzilla gives you faster cores in a smaller package, saving on rackspace and system power. As TPM carefully avoided mentioning in his "analysis", a Tukzilla 4-core will consume less power than two of the existing dual-cores when we just look at the bare chips alone, and as Tukzilla servers will require less sockets to make the same core count, that means they need less motherboards and therefore even less power than the previous generation of Integrity. I wonder why TPM didn't think to mention that?
RE: Always late and not worth the wait
Well, not much to say to this constant regurgitation of the same flawed IBM FUD. I suppose we can all guess where those awful Sun marketting people went to hide after the Sunset.
".....Itanium 9300 is a forklift upgrade. Their is no upgrade path from any prior chip....." Itanium in Integrity has used the same chassis and many server components since 2001, and the new generation looks like it will be good for not just Tulzilla but also Kitson and Poulsen. I'd say that's a lot better than Power. Every Power generation has required a forklift upgrade. As it looks now, the change to Power8 will require one too. Trolls that live in glass houses.....
"....Some glaring deficiencies are the 65nm fab and 1.73GHz....." <Yawn> Been here, debunked this before! Can't you get anything new to whine about? Please, it's getting boring, there's just no sport in it if this is the best you can do! Firstly, as I tried to explain before in terms a complete layperson might understand, clock isn't the only factor, which is why lower-clocked 9100-series Itanium can beat higher-clocked Power6, and no doubt why Tukzilla will cane Power7 in the same way. It's what is getting done with each clock tha matters, and part of that is keeping the cores spinning with data - how many of those fast clocks does Power spend sitting idle whilst the rest of the system tries to get data to the cores? Ever wonder why IBM had to design such massive cache into Power? And then rig TPC benchmarks so they could use a whole system's cache for one core, just to make Power look faster? DDR3 will maintain Itanium's advantage over Power in making better use of those clocks. Secondly, Intel have already proven a die process down to 25nm, which is way ahead of IBM's Power, and there is nothing now to stop Intel ramping down the die size on Tukzilla other than the fact that it will only be two years until we get the next gen. Sorry, but does the public Power roadmap say anything other than "Future" for Power8?
".....VMWare does not work on Itanium...." And since when has VMware run on Power? And with IVM, why would hp need VMware on Itanium?
"....RedHat RHEL6 - another ISV to drop support....." IBMers shouldn't mention Linux on Itanium, it only allows comparisons to Linux on Power. Like how Itanium has always trounced Power when it comes to Linux shipments, such as in 2005 when hp shipped five times as many Integrity servers with Linux as IBM did Power servers with Linux. And hp has maintained that advantage ever since. Whilst IBM brags about being multi-OS, their limited options and the limited will to support the customer going with Linux mean those customers prefer to buy their Linux servers from hp. I won't mention ProLiant either, as the story of hp's dominance of the x64 Linux market is even funnier!
"....I see SGI and Unisys have dropped out of the Itanium vendor list....." Whilst i can't comment on Unisys as I don't know anyone that works there, a friend at SGI would like to point out to you that SGI are still a member of the Itanium Solutions Alliance, and one of the small group of sponsor members. Now, time for a little IBM history that will probably really upset you.
Long, long ago, in the day when Ponytail was stil a semi-respected figure in the IT industry, and Sun was still mentioned without the automatic wince or laugh, a group of companies brought out their first Itaniums servers. Amongst that group was IBM with the x450, though many suspected IBM were still there more as a spoiler than as a real participant in the Itanium market. In fact, IBM salesgrunts only used the new IBM Itanium systems as a means to get into conversations they would otherwise have been excluded from, and then followed the standard IBM sales mantra - "try and convert the deal to a mainframe, if that doesn't work try selling a Power server and AIX, then if all else fails try selling Xseries, and don't mention things like Linux and Windows unless the customer does first" So, with all that determination to not sell, you'd think that IBM would never have sold a single Itanium server, so you may be susprised to know that IBM used to be the second largest Itanium server seller! Yes, those poor IBM salesdrones dug up 10,000+ instances where an Itanium server was the only solution, despite their desperate attempts to flog Power instead!
"....For a 2 Billion transistor processor and 24MB of cache you would think it could have more than 4 cores...." Well, I suppose that - seeing as you are obviously not a semiconductor engineer - that you could expect anything you like. The reality is Intel are still designing large and complex cores, whereas Power's cores are getting weaker with every generation. After spending years deriding EPIC and CMT, IBM are now desperately designing EPIC features into their RISC-based Power design and at the same time simplifying the cores as Sun tried with CMT, all so they can squash more cores onto a die. This desperate IBM design process is generating a mongrel chip with little hardware compatibility between generations and hence the constant need for forklift upgrades, and is driven by the fact Xeon is going 8-core, which is eating the Power base from below. In the long-run, as Sun found out, you can't keep simplifying the cores and still make it compete with a real enterprise CPU. The only way forward for IBM that will generate viable and progressive improvements in both core power and throughput is a complete and radical redesign of Power, and that will require a complete overhaul of the OS as well. Sounds expensive, and we all know where IBM likes recouping their expenses - from the customer.